1 / 14

CSCE 313: Embedded Systems Final Project

CSCE 313: Embedded Systems Final Project. Instructor: Jason D. Bakos. Final Project. Objective: Begin with Lab 5 Convert all floating-point computations to fixed-point Increase resolution to 1024 x 768 Use fixed-point representation for: z c min_x , max_x , min_y , max_y

fern
Download Presentation

CSCE 313: Embedded Systems Final Project

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. CSCE 313: Embedded SystemsFinal Project Instructor: Jason D. Bakos

  2. Final Project • Objective: • Begin with Lab 5 • Convert all floating-point computations to fixed-point • Increase resolution to 1024 x 768 • Use fixed-point representation for: • z • c • min_x, max_x, min_y, max_y • x2+y2 (for checking for divergence)

  3. Fixed-Point Review • Recall: fixed-point has fixed range • Recall: Range of non-fixed-point n-bit integer: • -2n-1 <= val <= 2n-1-1 • Range of signed (n,m) value: • -2n-m-1 <= val <= 2n-m-1 – 2-m • Need to decide where to set decimal point • For c, [min|max]_[x|y]: • Need to represent values from -2 to 2 • z should cover the worst case for a diverged pixel • x2+y2 should cover the worst case for the r2 of a diverged pixel

  4. Worst Case Analysis (0,2) (1.4,1.4) (2,0)

  5. Range and Precision • Problem: • Multiply (32,n) val with (32,m) val => (64,n+m) val • Need a way to get 64-bit product, or we lose the upper 32 bits

  6. Multiply • Assume 4-bit registers, need 8 bit product: x 10 11 11 10 = B1 B0 A1 A0 11 14 154 3*2=6 A0*B0 (4b) 0000 + 3*3=9 -> 9*4=36 A0*B1 (4b) 00 00 = A1*B0 (4b) 00 00 2*2=4 -> 4*4=16 + + 2*3=6 -> 6*16=96 A1*B1 (4b) 0000 154

  7. Example Fixed Point Multiply • Multiply A = (32,n) val andB = (32,n) val, need C = (32,n) product: • Declare A and B as “long” • Declare C as “long long” • Cast A and B as “long long”, • C = (long long)A * (long long)B; • Convert C from (64,2*n) to (32,n): • Shift C n bits to the right • Return C as int

  8. High Resolution • Goal: Increase resolution from 320x240 to 1024x768 • Problems: • Native resolution of VGA Controller is 640x480, so hardware modification is needed • SRAM is only 512 KB, not large enough to store a higher resolution frame, so we need to move pixel buffer to SDRAM

  9. Hardware Modification • Remove SRAM interface, connect DMA controller master interface directly to SDRAM • Remove rescaler (no longer needed) • Change DMA controller settings to 1024x768 • Each time you generate in SOPC Builder, must make edits to two generated Verilog files

  10. Hardware Modification • VGA Controller Verilog file, line 78 (after each generation): parameter CW = 9; parameter DW = 29; parameter R_UI = 29; parameter R_LI = 20; parameter G_UI = 19; parameter G_LI = 10; parameter B_UI = 9; parameter B_LI = 0; /* Number of pixels */ parameter H_ACTIVE = 640; parameter H_FRONT_PORCH = 16; parameter H_SYNC = 96; parameter H_BACK_PORCH = 48; parameter H_TOTAL = 800; /* Number of lines */ parameter V_ACTIVE = 480; parameter V_FRONT_PORCH = 10; parameter V_SYNC = 2; parameter V_BACK_PORCH = 33; parameter V_TOTAL = 525; parameter NUMBER_OF_BITS_FOR_LINES = 10; parameter LINE_COUNTER_INCREMENT = 10'h001; parameter NUMBER_OF_BITS_FOR_PIXELS = 10; parameter PIXEL_COUNTER_INCREMENT = 10'h001; parameter CW = 9; parameter DW = 29; parameter R_UI = 29; parameter R_LI = 20; parameter G_UI = 19; parameter G_LI = 10; parameter B_UI = 9; parameter B_LI = 0; /* Number of pixels */ parameter H_ACTIVE = 1024; parameter H_FRONT_PORCH = 24; parameter H_SYNC = 136; parameter H_BACK_PORCH = 160; parameter H_TOTAL = 1344; /* Number of lines */ parameter V_ACTIVE = 768; parameter V_FRONT_PORCH = 3; parameter V_SYNC = 6; parameter V_BACK_PORCH = 29; parameter V_TOTAL = 806; parameter NUMBER_OF_BITS_FOR_LINES= 11; parameter LINE_COUNTER_INCREMENT = 11'h001; parameter NUMBER_OF_BITS_FOR_PIXELS = 11; parameter PIXEL_COUNTER_INCREMENT = 11'h001;

  11. Hardware Modification • Clocks Verilog file, line 69 (after each generation): parameter SYS_CLK_MULT = 1; • Change to: parameter SYS_CLK_MULT = 2; • Line 174: DE_Clock_Generator_System.clk2_divide_by = 2, DE_Clock_Generator_System.clk2_duty_cycle = 50, DE_Clock_Generator_System.clk2_multiply_by = 1, • Change to: DE_Clock_Generator_System.clk2_divide_by = 7, DE_Clock_Generator_System.clk2_duty_cycle = 50, DE_Clock_Generator_System.clk2_multiply_by = 9,

  12. Software Modification • New pixel buffer requires 1024x768x2 = 2359296 bytes • Allocate in CPU0 • Give 3.5 MB to CPU0, 1.5MB each for CPUs 1-3 • Allocate in SDRAM (as a global variable): #ifdef MASTER // define this in Eclipse project settings for CPU0 volatile alt_u8 pixel_buffer_memory[2359296]; // 768x1024x3 #endif • Add –DMASTER in cpu0’s project settings under Nios II Application Properties

  13. #define

  14. Software Modifications • Copy address of only one processor’s buffer to the DMA controller for F/B buffer (in main()): alt_up_pixel_buffer_dma_change_back_buffer_address(my_pixel_buffer,(unsigned int)pixel_buffer_memory); alt_up_pixel_buffer_dma_swap_buffers(my_pixel_buffer); while (alt_up_pixel_buffer_dma_check_swap_buffers_status(my_pixel_buffer)); alt_up_pixel_buffer_dma_change_back_buffer_address(my_pixel_buffer,(unsigned int)pixel_buffer_memory); • Subsequent writes to buffer using address stored in DMA controller will write the same buffer • This is because pointer is stored in global location • Avoid divide: add a fixed-point constant for (1/768) and (1/1024) and multiply these

More Related